The Agent Skills Directory

[COMMAND_EXECUTION]: The skill constructs and executes a shell command using the demucs command-line utility. It passes user-provided inputs such as file paths, model types, and device configurations as arguments to the tool.
[EXTERNAL_DOWNLOADS]: Upon the first execution, the tool fetches approximately 2GB of pre-trained model weights from Meta Research's official infrastructure. This is standard behavior for machine learning models of this nature.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-supplied audio files and directory paths. While it lacks explicit boundary markers or sanitization logic within the instructions, it relies on signal processing tools from a trusted vendor, which significantly limits the potential for instruction-based attacks through binary data.

separate-stems